• Cuda Occupancy Calculator External Link. Use this to see how many threads/warps will run on your GPU at any given point of time. You need to specify:

1. The Compute Capability: 1.3 for GTX280
2. Threads/block: You decide this
3. Registers per thread: Use the nvcc compiler option -ptxas-options=”-v“. It will report the number of registers per thread.
4. Shared memory per block: see the item above. Again you'll have to use nvcc's output.

• Programming Massively Parallel Processors: A Hands-on Approach, David Kirk and Wen-mei Hwu, External Link.
• Cuda by Example, Jason Sanders and Edward Kandrot, External Link.
• The OpenCL Programming Book, Ryoji Tsuchiyama, Takashi Nakamura, Takuro Iizuka, Akihiro Asahara, Satoshi Miki, External Link

We will be using CUDA v3.2. A newer *candidate* version exists today (v4.0 RC). The lab machines will not be upgraded for the duration of this course as we want to stick with a stable version.

• You can download all reference material and tools from here
• CUDA Programming Guide, External Link
• CUDA Best Practices Guide, External Link
• CUDA Toolkit Reference Guide, External Link
• More reference guides from NVidia (CUDA and OpenCL). Look at the Linux section and at the CUDA Tookit row External Link

• What Every Computer Scientist Should Know About Floating-Point Arithmetic local copy, External Link